Research On Fast CU Partition Algorithm Based On H.266/VVC

With the development and improvement of the H.266/VVC coding standard,the improvement of video processing efficiency has driven the development of the video industry,which lays a foundation for the development of a new generation video coding standard.To satisfy the requirements of people for the definition,fluency and real-time of video,a new generation video coding standard came into being,which has good network adaptability,parallel processing capability and compression efficiency and will be well applied in many fields.Consequently,H.266/VVC video coding standard is a hot spot of current research.The quad-tree with nested multi-type tree(QTMT)partition structure introduced by H.266/VVC greatly improves the coding efficiency,but leads to a highly coding computational complexity at the same time.This paper proposes two effective solutions for this problem:(1)A fast Coding Unit(CU)partition algorithm based on Random Forest Classifier(RFC)is proposed.According to the texture complexity,the CU is divided into smooth,ordinary,and complex region.The ordinary region utilizes the original intra coding process for coding prediction,the smooth area is no longer divided,and the CU in the complex region uses the pre-trained RFC for classification.Experimental results show that compared with the original encoder of H.266/VVC,this algorithm can save about 39.05% coding time,and BDBR is0.73%,which is negligible.(2)A fast CU splitting algorithm based on an improved Directed Acyclic Graph-Support Vector Machine(DAG-SVM)classifier is proposed.The improved F-score feature selection method is used to select the features related to the CU partition for obtaining the feature subset.The center of gravity of each class is firstly calculated when constructing the improved DAG-SVM classifier,and the distance between each class is calculated by the center of gravity.Then,the average value of the distance between each class and other classes are computed,where the class with a larger average value is farther away from other classes.The class with a larger average distance is preferred to generate the upper classifier in the Directed Acyclic Graph(DAG).Finally,the trained DAG-SVM classifier predicts the best CU partition in advance.Experimental results show that compared with the original encoder of H.266/VVC,this algorithm can save about 54.74% coding time and BDBR is only 0.92%.This paper proposes an algorithm for reducing the coding complexity of the CU partition process in H.266/VVC,which has effectively reduced the coding time and promoted the practical application of H.266/VVC.